SLH: Introduction to Bayesian Optimization and Its Applications in Transportation and AI

This talk introduces Bayesian Optimization (BO), a sample-efficient framework for optimizing expensive, noisy black-box functions. We will cover the core ideas behind surrogate modeling (e.g., Gaussian processes) and acquisition functions (such as EI and UCB) that balance exploration and exploitation to find high-performing solutions with limited evaluations. The applications focus on two representative directions: (1) parameter calibration for transportation simulation—tuning behavioral and network parameters so simulated traffic patterns match real observations; and (2) hyperparameter optimization in machine learning—automatically selecting model and training settings to improve accuracy, robustness, and efficiency. We will also highlight practical considerations such as constraints, multi-objective trade-offs, and scalable implementations.

Presented by NYU’s Yu Tang